fMRI Study Design

The basic framework

Russell James, J.D., Ph.D.Texas Tech University

Goals of fMRI Study Design

1. Create a desired

cognitive state

(standard experimental design issues)

2. Detect brain signals

associated with that state

(fMRI-specific experimental design issues)

We want to estimate the likelihood that a voxel, or group of voxels, is

responding to the stimulus

But, fMRI data is not like this

Activation

fMRI data is like this

Activation

The signal change is small.The signal is noisy.

Activation

The signal

change is small

½% to 3% change in intensity with a 1.5 T scanner

The signal is noisy

1. The brain is noisy

2. The scanner is noisy

The brain is constantly active, constantly firing, constantly receiving input, constantly sending instructions

The brain is noisy

Even conscious thought is scattered. Did you think about something other than fMRI in the last 3 minutes?

The brain is noisy

1. Contrasts 2. Repetition

How do we

design for noisy brains?

Think in contrasts

Task A Task BTask A-Task B

A single image contains much

unrelated brain activations

A contrast can subtract out

the noise

Think of study design in terms of contrasts

Image of task

A

Image of task

B

Image of task A-

Image of task B

The comparison task can be “rest”

Image of task

A

Image of task

B

Image of task A-

Image of task B

We can use a “cognitive subtraction”

comparison to isolate an activity

- =

Cognitive subtraction: the comparison task is

identical, except for one variation of interest

Cognitive subtraction: View famous faces v. non-famous faces

How might you improve cognitive subtraction in this picture selection?

Match gender? Color pallet? Expressions? Clothing choice?

Levels of cognitive subtraction

• Basic cognitive subtraction: Rational argument of validity

• Cognitive conjunction: Two cognitive subtraction designs show same activation difference

• Parametric design: Increasing levels of a factor correspond with increasing levels of activation

Stron

ger Resu

lts

Cognitive conjunction

Memorizingnumbers

Memorizingletters

Reading numbers

v.

v.

Parametric design

Memorize 9 numbers

Memorize 6 numbers

Memorize 3 numbers

Memorize 1 number Incr

easi

ng

acti

vati

on

● ●

● ●

● ●

Design for BIG contrasts

fMRI is too noisy to be good at subtle cognitive differences

Contrast extreme differences first

1. Contrasts 2. Repetition

How do we

design for noisy brains?

Design for repeated activations

In fMRI we get 25 people to do a task 40 times

for 1,000 activations

In social sciences we may get 1,000 people to make one

decision on a survey

Block Design

Two common design types

Event Related Design

Block Design

Time

Within a certain range, repeating a stimulus will cause the HRF to stack linearly.

This causes large total signal change in block designs.

Stacking the HRFs

Event Related Design

• Easiest way to detect differences among two comparison states

• Largest HRF activation

• More robust to unexpected HRF shapes

Block design advantages

BOLD response in block v. event related (slow)

• Modeling incorrect responses (can’t know in advance)

• Habituation may prevent activations

Can precisely observe the actual HRF

Permits self-paced trials

Event related design advantages

Block designs may not work

Barriers to repeating the activation

Barriers to repeating the activation

• Repetitions can get boring or predictable, reducing activation

• Emotional states may not be induced or changed quickly

• Some decisions are difficult to repeat

• Some biases can be added once, but not removed

Repetitions can get boring or predictable, reducing activation

+ + +

+ + +

+ + +

Emotional states may not be induced or changed quickly

Think of the happiest moment of your life for 12 seconds

Think of the most painful moment for 12 seconds

Think of the 2nd happiest moment for 12 seconds

Think of the 2nd most painful moment for 12 seconds

Some decisions are difficult to repeat

“If you died today what percentage of your estate would you want to leave to your children?”

Are you sure?

Any second thoughts?

Want to think about it some more?

Some biases are not easily removed

1. How much of a $100 extra payment will you give to the United Way?

2. The United Way CEO made $1,037,140 last year.

3. How much of a $100 extra payment will you give to the United Way?

1st comparison works great, but can you repeat this?

The signal is noisy

1. The brain is noisy

2. The scanner is noisy

Designing for a noisy scanner

Physical issues

Timing issues

Physical issues of a noisy scanner

Machine Metal Movement

Machine

Optimizing machine, setting

parameters, region of

interest, speed v. resolution, etc.

Metal: Limit any ferrous content to avoid disturbances in magnetic field

Movement

• Motivated subjects

• Foam packing

Statistical adjustments (but, may not help task-related movement)

Keep the subject still during runs

Designing for a noisy scanner

Physical issues

Timing issues

Timing issues of a noisy

scanner signal

Scanner Drift

Over longer periods of time (2-10 minutes), the magnetic field of the scanner can slowly rise and fall.

Scanner Drift

Comparison across long (>2min) periods should be avoided.

Condition A Condition A

Condition B Condition B

A–B shows magnetic differences but not from HRF

What is wrong with this?

30 sec block of task A (version 1)30 sec block of task A (version 2)30 sec block of task A (version 3)30 sec block of task A (version 4)30 sec block of task B (version 1)30 sec block of task B (version 2)30 sec block of task B (version 3)30 sec block of task B (version 4)30 sec block of task C (version 1)30 sec block of task C (version 2)30 sec block of task C (version 3)30 sec block of task C (version 4)

What is wrong with this?

30 sec block of task A (version 1)30 sec block of task A (version 2)30 sec block of task A (version 3)30 sec block of task A (version 4)30 sec block of task B (version 1)30 sec block of task B (version 2)30 sec block of task B (version 3)30 sec block of task B (version 4)30 sec block of task C (version 1)30 sec block of task C (version 2)30 sec block of task C (version 3)30 sec block of task C (version 4)

Comparing A to C spans well over 120 seconds so we can’t distinguish from scanner drift

Block Design Timing Issues

Ideal timing 15-20 seconds on then 15-20 seconds off (or A then B)

• Long enough for HRF to relax in between presentations

• Short enough for many comparison blocks within short time

Slow Event-Related Timing Issues

Waiting 12+ seconds in between each event to allow HRF to calm down

Boring and inefficient

Rapid Event-Related DesignTiming Issues

Jitter spacing to record different parts of the HRF and to avoid correlation with other functions like heartbeat and breathing

Rapid Event-Related DesignTiming Issues

Gap spacing >4 seconds, else• HRF blurring: Not enough time

for noticeable HRF changes • Non-linearity: HRFs don’t stack

linearly foreverOptimum jittering estimation programs (e.g., OptSeq - Doug Greves; Genetic Algorithm - Tor Wager)

Session Timing

Typically, studies include groups of tasks of 4-10 min. with intervening 2 min. breaks.

Also need high resolution (T1) scan ~5 min.; Locater scans (~30 sec.); T2 axial scan for radiologist (~2 min.)

10 min. subject in and out

Locator – 20 seconds

Short Break

T2 axial: 2 min

Break

T1 high resolution: 5 min

Break - wake up

Block 1: 8 minutes

Break/Instructions: 2 min

Block 2: 8 minutes

Break/Instructions: 2 min

In a 48 minute session, you may get about 24-28 minutes of actual stimulus presentation

Block 3: 8 minutes

$650/hour scanner time ~$1300/hour stimulus time~$20+ per minute~$1+ every 3 seconds

Add 12 sec. to 25 subjects = $100+

With so many, many design issues to think about, what is

the best way to design your first study?

Stand on some shoulders!

Find a good prior study, copy the technical elements, but change your item of interest

fMRI Study Design

The basic framework

Russell James, J.D., Ph.D.Texas Tech University